Recognizing upper limb movements with wrist worn inertial sensors using k-means clustering classification

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Pattern Recognition

Background:

• Assessing neurodegenerative disease recovery, such as stroke or cerebral palsy, requires objective monitoring of patient movement.
• Current methods may lack continuous, real-world data capture for specific limb function.

Purpose of the Study:

• To develop and validate a pattern recognition methodology using a single wrist-worn inertial sensor.
• To detect three fundamental human forearm movements (extension, flexion, rotation) during daily activities.
• To evaluate the potential of this technique as a clinical tool for rehabilitation progress assessment.

Main Methods:

• Utilized data from a tri-axial accelerometer and gyroscope in a wrist-worn sensor.
• Employed a training phase with feature selection (sequential forward selection) and k-means clustering.
• Applied minimum distance classifiers (Euclidean/Mahalanobis) for movement detection during a 'making-a-cup-of-tea' activity.
• Compared results with Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) classifiers.

Main Results:

• Achieved an average accuracy of 88% (accelerometer) and 83% (gyroscope) in healthy subjects.
• Demonstrated an average accuracy of 70% (accelerometer) and 66% (gyroscope) in stroke survivors.
• Successfully detected forearm extension, flexion, and rotation during a simulated daily activity.

Conclusions:

• The proposed wrist-worn sensor and pattern recognition methodology effectively detects fundamental forearm movements.
• This technique shows potential as a non-invasive, objective tool for monitoring rehabilitation progress in neurodegenerative conditions.
• Further validation in clinical settings is warranted to establish its utility in patient care.